Machine learning in trading: theory, models, practice and algo-trading - page 3458

New comment
 
Maxim Dmitrievsky #:
Need
And here we go again with this unhealthy topic about some events.

Pile the signs into the bousting, pick out what it predicts well, cluster it into groups. There are your patterns. How much more?

What are you using?

Busting looks for patterns, i.e. some sequences that have some generalisability. Nothing like mytarmailS does.

 
The most interesting topic to pursue is converting the source series to stationary with the least loss of information. Then there will be no need to mine anything.

For example, increments levelling information about price levels. And the original series cannot be put into the model. Fractional differentiation also saves little memory.

Visually, it should look like a quotir in flat, but not normally distributed, but uniformly distributed, then the classifier will predict like a baby.
 
Maxim Dmitrievsky #:
transformation of the original series to a stationary one with the least loss of information.

For example, increments levelling information about price levels.

The increments contain exactly as much information as the initial prices. Since the two series are derived unambiguously from each other.


Strangely, it is better for the MO to feed someone (not the MO itself) the increments transformed from prices as input.

 
fxsaber #:
The increments contain exactly as much information as the initial prices. Because these two series are obtained unambiguously from each other.
Not so much if the trend is subtracted from the original series
 
Maxim Dmitrievsky #:
Not the same amount if you subtract the trend from the original series

You're inputting data that is convenient for MO. A lousy MO then, or notions of it.

A human being is always given prices as input. And he has to help the MO with some transformations.

 
fxsaber #:

You're inputting data that is convenient for the MoD. A lousy MO then, or ideas about it.

A human being is always given prices as input. And he has to help the MO with some transformations.

I'm talking about loss of information during transformations. It's not about the lousiness of the IO or any idea of it. There was even an article about fractional differentiation.
 
Maxim Dmitrievsky #:
It's about loss of information during transformations. There was even an article about fractional differentiation.

The point is that you invent transformations specifically for the MO. And not the MO creates convenient for itself transformations from the initial data.

 
fxsaber #:

The point is that you come up with the transformations specifically for the MoD. It is not the MoD that creates convenient for itself transformations from the source data.

No, the point is that transformations remove information about price levels from quotes.
 
Maxim Dmitrievsky #:
No, it's about conversions removing price level information from quotes.

You do transformations for some reason. And obviously you are not doing it to look at them yourself. You're converting them to feed them to the IO.

You are forced to invent stationary transformations with information loss, because you know that the MO copes better with stationary data.
 
fxsaber #:

You're doing the conversions for a reason. And obviously you're not doing it to look at them yourself. You're converting it to feed it to the MoD.

Of course
1...345134523453345434553456345734583459346034613462346334643465...3516
New comment
Reason: